Summary:
As one of the most active fields within energy companies, financial forecasting always looks for the latest research improvement to improve its accuracy. Whether used to analyse trend lines, hedge futures positions, maximise profit or better risk management, such models are now located at the heart of multiple trading strategies. With the development of data availability and computational power, Machine Learning and Deep Learning models are now being implemented alongside existing quantitative and econometrics models. In this research, the forecast of jet fuel prices with multiples forecasting models yielded different outcomes. A Long Short-Term Memory (LSTM) neural network and a Support Vector Regression (SVR) model was constructed and compared to the most popular forecast model, the ARIMA. Based on three different evaluation metrics, the LSTM model outperforms any other model. Furthermore, since it relied on the training of large datasets and optimised parameters, the model produced a unique forecast that focuses on past observations to predict price movement. On the other side, the SVR performed better for shorter predictions with less data availability. This research was designed to understand better the working of such models whilst understanding how their strengths and weaknesses could benefit forecasting analysis.
If interested in the paperwork or discuss further about the findings, please contact me at: romain.boucaumont@gmail.com